Description
Sonnet 5 vs Opus 4.8, tested inside my real 168 GB folder with a team of 50 AI agents. The result was not what most people expect.
👉 Join for free and build your own AI productivity system: https://www.myicor.com/?utm_source=youtube&utm_medium=video&utm_campaign=XlKVo5wsl8Y&utm_content=description
Here is the honest takeaway. At the effort level I actually need for real work, Sonnet 5 ends up costing MORE than Opus 4.8, not less. But the deeper truth from this benchmark is bigger than any single model. It is not the model that makes the results, it is the FOLDER. The scaffold, the instructions, the guidelines, and the team of agents are what pull the most out of Claude Code, whether you run Sonnet 5, Opus 4.8, or even a cheap open source model.
I run both models side by side in Claude Code, give them the same real task inside my personal folder, and compare speed, token cost, identity pickup, and how much of the folder each model actually reads. Then I break down the pricing chart that has everyone confused about why Sonnet 5 can cost more than Opus 4.8.
CHAPTERS
0:00 Testing Sonnet 5 vs Opus 4.8 in a real 168 GB folder
0:32 Why I skip Claude Desktop and Cowork
1:13 The folder that guides the model: a team of 50 agents
2:04 Starting the Sonnet vs Opus head to head
3:52 Running both models side by side
4:46 The Who Are You test: fast vs comprehensive
5:35 The real task: summarize everything shipped for myICOR
5:55 Opus becomes Larry and delegates, Sonnet spins a generic agent
8:55 Why spending more tokens upfront can be cheaper
9:33 Running open source GLM 5.2 in the same folder
10:52 Persistent memory: session logs and team journals
14:24 Comparing the two HTML reports
17:02 Asking Claude to compare both reports
18:59 The result: Opus read the whole folder, Sonnet did not
20:15 The pricing chart: why Sonnet 5 costs more than Opus 4.8
22:02 It is not the model, it is the folder
22:51 Get the scaffold free inside myICOR
23:15 Your turn: did you test Sonnet 5?
RESOURCES
🚀 Join myICOR free and start building your AI productivity foundations: https://www.myicor.com/?utm_source=youtube&utm_medium=video&utm_campaign=XlKVo5wsl8Y&utm_content=description
✨ Connect your AI to your myICOR account via the MCP Server, so Claude gets full context about your tools and your progress.
📂 Free ICOR® Starter Kit (the scaffold from this video): https://plmv.world/free-icor-starter-kit
Tools we use at the Paperless Movement:
🎙️ Voice Dictation, Wispr Flow: https://ref.wisprflow.ai/paperless-movement
🧠 PKM Deep Thinking, Heptabase: https://plmv.world/heptabase
🌊 PKM Shallow Thinking, Tana: https://plmv.world/tana
✅ Task Management, Todoist: https://plmv.world/todoist
⏰ Time Management, Sunsama: https://www.sunsama.com/
🚀 Project Management, ClickUp: https://plmv.world/clickup
👥 Team Deep Thinking, Miro: https://plmv.world/miro
📧 Email Management, Superhuman: https://plmv.world/superhuman
🤖 AI Meeting Notes, MeetGeek: https://plmv.world/meetgeek
💻 MacOS Productivity Boost, Raycast: https://plmv.world/raycast
🎥 Our primary video editing software is Descript: https://plmv.world/descript
----------
Follow me
• Productivity for Professionals (Systems building Channel): https://www.youtube.com/@prodforpros
• Think. Map. Execute. (Tools Channel): https://www.youtube.com/@myicorstories
• Productivity like a Pro (Podcast Channel): https://www.youtube.com/@productivitylikeapro
• Tom at Paperless Movement (This Channel): https://www.youtube.com/@paperlessmove
• Linkedin: https://www.linkedin.com/in/tomsolid/
Follow Paperless Movement®
• Inside our free membership: https://www.myicor.com/?utm_source=youtube&utm_medium=video&utm_campaign=XlKVo5wsl8Y&utm_content=description
• Linkedin: https://www.linkedin.com/company/paperlessmove/
Disclaimer: Some of the links in this description may be affiliate links. This means that if you click on one of the links and make a purchase, we may receive a small commission at no additional cost to you.
----------
#ClaudeCode #Sonnet5 #ClaudeAI #Opus4 #AIProductivity #Anthropic #myICOR #AITools #ProductivitySystem #AIagents